SPATIO-TEMPORAL TRENDS OF INFANT MORTALITY IN BRAZIL
Marcia C. Castro
Department of Global Health and Population
Harvard School of Public Health
Boston, MA, USA
[email protected]
Celso C. S. Simões
Department of Social Indicators
Brazilian Institute of Geography and Statistics (IBGE)
Rio de Janeiro, RJ, Brazil
[email protected]
August, 2009
Paper prepared for presentation at the XXVI IUSSP International Population
Conference, Marrakech, Morocco, September 2009.
INTRODUCTION
Infant mortality in Brazil has declined appreciably in the past decades, with most significant
gains observed since the 1970s (IBGE 1999; Simões 2002). Between 1930 and 1970, the infant
mortality rate (IMR) for the country declined by 29.2%, but still remained above 100 per 1,000
live births. Between 1970 and 2005, however, IMR was reduced by 79.7%. These gains
significantly contributed to increases in the country’s life expectancy at birth (Castro 2001), and
put Brazil in a favorable position to achieve internationally set goals to reduce mortality among
infants. In 1990, the World Summit for Children adopted a target to reduce the 1990 IMR by one
third, or to 70 per 1,000, whichever was the greater reduction, by the year 2000 (UNICEF 1990).
Brazil registered a 42% decline in IMR between 1990 and 2000, surpassing the set target, and
achieving an IMR of 27.6 per 1,000 in 2000. Also, the decline in infant mortality places Brazil
among the few countries likely to meet the Millennium Development Goal (MDG) to reduce
child mortality by two thirds between 1990 and 2015 (UNICEF 2005).
Nonetheless, with an IMR of 23.5 per 1,000 in 2005, Brazil still lags behind other countries
in South America, such as Uruguay (13.1 per 1,000) and Chile (7.2 per 1,000). Most importantly,
the overall decline in IMR does not necessarily indicate that inequalities in infant mortality
within the country were equally reduced. This is a critical issue for Brazil, since the country has
historically ranked high in terms of income concentration worldwide – Brazil ranked 2nd in 1998
(World Bank 1999), and in 1999 it was the country with the highest ratio between the average
income of the 20% richer and the 20% poorer, above 30 (Barros, Henriques and Mendonça
2000). Despite recent improvements (Arbix 2007), the Gini Index in 2007 was at 57, placing
Brazil as the 11th most unequal country in the world (World Bank 2007).
2
Regional inequality in Brazil regarding access to services, as well as social and health
outcomes has been widely recognized: while the South and Southeast regions regularly report
higher income and better socio-economic indicators, the North and Northeast regions often lag
behind (Hoffmann 2000; Medici and Agune 1994; Rocha 1998). As a result, IMRs present a
distinct clustering pattern across micro-regions, with high rates often concentrated in the
Northeast, and low rates in the Southern portion of the country. Despite this pattern, declines in
IMR have occurred countrywide. In an ideal scenario, these declines should be accompanied by
reductions in inequalities between rich and poor areas. However, the “inverse equity hypothesis”
(Victora et al. 2000) postulates that inequalities may temporarily increase or remain unchanged
after the introduction of interventions targeted to reduce infant mortality. The increase is
expected to hold until richer areas have reached low levels of IMR (with further improvements
harder to achieve), and poorer areas achieve wider access to means for promoting policies
toward IMR decline (Victora et al. 2000).
In this paper we address the issues discussed above, and aim to investigate spatio-temporal
trends in infant mortality in Brazil between 1980 and 2005, utilizing a time series of indirectly
estimated IMRs for micro-regions (Simões 1999). Specifically, we have three major goals. First,
we appraise the declines in IMR and in inequalities in infant mortality, in light of the “inverse
equity hypothesis”. Second, we evaluate the presence of clusters of high and low infant mortality
in micro-regions, and assess if/how the clustering pattern changed over time and across space.
We also investigate the clustering pattern of the relative change in IMR for 5-year periods of
time in an effort to examine whether or not the gains were primarily concentrated in specific
areas of the country. Third, we construct a multilevel model for change, also known as random
coefficients model (Singer and Willett 2002), as an initial attempt to understand how trajectories
of IMR decline differ within and between micro-regions in Brazil.
3
The remainder of this article is divided into 5 sections. The next section provides a brief
historical background on socioeconomic, political, and demographic changes in Brazil potentially
related to trends in infant mortality. In the second section we describe the data utilized in the
analysis. The methods chosen for IMR estimation, evaluation of spatial clustering, assessment of
inequalities, and construction of a change model are detailed in the third section. Our findings
are summarized in section four, and we conclude with a discussion in the final section.
HISTORICAL BACKGROUND
Gradual declines in infant mortality were observed in Brazil since the early 20th century, but
IMR levels remained high until the 1990s, when the country registered, for the first time, an
infant mortality lower than 50 per 1,000. Nonetheless, within-variation among administrative
regions was substantial, with the Northeast region registering mortality rates 2.6 times higher
than the rates recorded in the South region in 1990. Next, we focus on the period post 1930s to
review important political, social, economic, and demographic transformations likely to be
connected (directly and/or indirectly) with these trends in infant mortality.
The early 1930s are historically linked to the initial industrial expansion in Brazil (Cano
1988), which fostered the growth of the urban labor market, resulting in intense rural-urban
migration. In the same decade, studies focusing on hunger and nutrition commenced in Brazil,
culminating with the seminal work from Josué de Castro in 1946, who produced the first detailed
map of hunger in the country (Castro 1946). During the 1940s and 1950s the number of new jobs
opened in urban areas largely surpassed new offers in rural areas, progressively intensifying the
rural exodus. The country entered a transition phase: from an agrarian to an industrial society.
Rural-urban migration started to reshape the spatial distribution of the population, peaking in the
4
1960s, and resulting in accelerated and disorganized urban growth, concentrated in major centers
such as Rio de Janeiro (Brazil’s capital until 1960) and São Paulo, both located in the Southeast
region (Simões 2006). While Brazil was experiencing fast urbanization and industrialization,
mortality resulting from nutritional deficiency was still high. This contrast brought the discussion
of nutrition programs to the governmental agenda (Vasconcelos 2005).
The lack of policies to address the intense migratory flow, the precarious infrastructure in the
cities, the absence of an agrarian reform, and the deterioration of urban salaries became exposed
during the early 1960s. At that time, IMR deteriorated in the most developed regions of the
country (Southeast and South), exactly those where the urban expansion was mostly significant.
An economic recession occurred between 1962 and 1967, followed by a period of intense
economic growth (1968-1974). After the political transition to a military government in 1964,
important investments in education and habitation started to take place, but tough wage
adjustments were also imposed (Demo 1981). During 1965-1970, infant mortality observed the
lowest decline in Brazil since the 1930s, 3%. Income inequality became an important topic of
discussion in the early 1970s, highlighting the deterioration of health conditions of those
excluded from the benefits of intense economic growth. In the mid-1970s, 46% of children
younger than 5 years had nutritional deficiencies (Vasconcelos 2005). As a response to this and
many other challenges, the government established the Social Development Council in 1974,
directly connected to the President’s office, which was in charge of supporting social policy
making (Pinto 2002). Massive social programs with large national coverage were implemented
in Brazil in the 1970s, including, among others, initiatives focused on drug distribution to low
income population, special assistance to rural workers, promotion of social development in the
Northeast region, improvement of housing financing to population with low income, public
health interventions to reduce the burden of preventable diseases in the Northeast and in the
5
Amazon, improvement of epidemiological surveillance, specific actions on nutrition and
maternal and child health, and the implementation of the national health system. An initiative of
such magnitude was never seen before in Latin America (Simões 1999; Vasconcelos 2005), and
these programs had an overall positive impact in social indicators such as education, nutrition,
sanitation, vaccination coverage, and access to primary health care, contributing to declines in
infant mortality (Simões 1997).
During the 1980s, Brazil was facing a serious economic recession and fiscal crisis (Silva
1992). Development programs were starting to weaken, and structural adjustment programs
imposed restrictions on the availability of resources for social programs (Simões and Ortiz 1988;
Tavares 1981). A new political era started in 1985, with the end of the military regime. After
several economic plans, stability was achieved in the mid-1990s (Giambiagi et al. 2005; Pereira
1994), and new programs specifically targeted to reduce hunger, unemployment and extreme
poverty were launched. In 1995, Brazil implemented a conditional cash transfer (CCT) program
(the first in Latin America), targeting poor families and conditioned on school enrollment and
attendance of all children in school age (Pero and Szerman 2005). Other programs followed,
aimed at providing resources to improve access to basic health services and nutrition (Draibe
2003; Santos and Santos 2007; Vasconcelos 2005). The trend continued in the 2000s, focusing
on reduction of inequalities and poverty alleviation, and culminated with the implementation of
the “Bolsa Família” conditional cash transfer program in 2003, which unified four pre-existing
cash transfer programs, and targeted families below a defined poverty line, conditioned on
compliance with requirements to attend health checks and maintain school-age children in school
(Marques 2005). Currently, the “Bolsa Família” is the largest conditional cash transfer program
in the developing world (Lindert et al. 2007), with more than 11.5 million families covered until
June 2009 (approximately 48 million people, or 25% of the current population).
6
Specifically related to health, the Family Health Program, implemented in 1994, aimed at
improving access to primary health care, utilizing a community-based approach for local care
provision. Health care services are provided by a team comprised of at least one physician, one
nurse, one nurse assistant, and up to six community health workers; some teams may also
include a dentist and two assistants. Each team is responsible to provide care for up to 1,000
families (or approximately 4,500 people) in a determined geographical area. (Ministério da
Saúde 2009b). Other initiatives included special programs tailored to children with physical and
mental disabilities, promotion of breastfeeding, expansion of immunization coverage, and
control of vitamin A deficiency. A milestone was the implementation of the HIV/AIDS control
program, internationally recognized for it success (Berkman et al. 2005; Nunn et al. 2009).
Also important was the special focus devoted to all 1,133 municipalities (or 20.3% of
Brazil’s municipalities) that comprise the semi-arid region (Figure 1), which extends throughout
the Northeast region and parts of the state of Minas Gerais in the Southeast (Ministério da
Integração Nacional 2005), and has a population of approximately 20 million. The region is
prone to severe droughts since rainfall is concentrated in a short period of 3 to 4 months. During
the drought period women and children are forced to walk long distances in search for water.
Special initiatives implemented during 2001 and 2003 allowed the construction of water tanks
with capacity for 16 thousand liters of water, which capture and store rain water to be used
during the months of drought. These initiatives were scaled-up into a new governmental program
(“Cisternas”) to support the construction of water tanks in the semi-arid region, targeting families
living in rural areas and eligible to the “Bolsa Família” CCT program (Brito, Silva and d’Alva 2007).
Demographically, Brazil experienced high population growth between 1940 and 1960, an
average of 2.8% per year. After a similar growth between 1960 and 1970, the growth started to
7
decline in the 1970s (2.5%), reaching an average of 1.6% per year during 1991 and 2000
(Simões 2006). The total fertility rate (TFR) remained at high and relatively constant levels
between 1940 and 1960 (Carvalho 1980). A modest and slow decline since the early 1900s and
small oscillations in fertility in the 1950s and 1960s have been reported (Frias and Carvalho
1994; Simões 2006). Nevertheless, important demographic transformations started in the mid1960s. In four decades the TFR experienced a dramatic decline: from 6.3 in 1960 to 2.3 in 2000
(Carvalho 1997; Simões 2006). The National Demographic and Health Survey on Women and
Children, conducted in 2006, reported a TFR of 1.8, based on the birth history reported for 36
months prior to the interview (Ministério da Saúde 2008). Although fertility declines occurred in
all regions and across different socioeconomic groups, the North and Northeast regions had the
highest TFRs in 2000, 3.2 and 2.7, respectively, and their transition process was lagged by
approximately 10 years compared with other regions (Simões 2006). Fertility declines had
important impacts in infant mortality both through reduction in parity and increase in the interval
between births (Rutstein 2000).
DATA
All data gathered for this study are detailed by micro-region, which is an areal aggregation of
several municipalities (Lima et al. 2002). Brazil is currently divided into 5,565 municipalities,
557 micro-regions, and 5 regions (Figure 1), as defined by the Brazilian Institute of Geography
and Statistics (IBGE). We chose the micro-region as the spatial unit of analysis in order to
increase the robustness of the indirect estimations of IMR. Only three changes in the boundaries
of micro-regions were observed during the study period, and were resolved by reassigning
municipalities according to the current definition.
8
Information on deaths can be obtained through varied data sources in Brazil, but the
occurrence of underreporting, especially among children younger than 1 year in rural areas of the
North and Northeast regions, is still an impediment to calculate IMR directly. Therefore, indirect
techniques of estimation are needed, which demand the use of data only available in
Demographic Census and special surveys. In this study we used the Demographic Census data
(1980-2000) for the estimation of IMR by micro-regions. We also utilized the National Sampling
Household Survey (PNAD) collected in 2004, 2005 and 2006, combined with Census data
(1940-2000), to produce IMR estimates by regions. From each source we extracted information
on: (i) number of women aged between 15-49 years by 5-year age groups; (ii) number of
children ever born classified by age group of mother; and (iii) number of children surviving
classified by age group of mother.
Demographic Censuses were also utilized to construct a time series of urbanization
(percentage of the population living in urban areas) by micro-region. Data for intercensal years
were estimated utilizing the ai bi procedure, a mathematical model that allows the decomposition
of estimates for larger areas into smaller areas, assuring that the sum of the smaller areas
reproduce the initial estimated total for larger areas (Madeira and Simões 1972). Population
estimations (total and 0-1 year) for each micro-region were obtained at the Ministry of Health
on-line database (www.datasus.gov.br). The data combine Census information and annual
estimates produced by IBGE. Data on municipal gross domestic product (GDP) were available
through the Institute of Applied Economic Research (IPEA – www.ipeadata.gov.br) for 1980,
1985, 1996, and 1999-2006. All GDP data were expressed in constant 2000 Reais (the Brazilian
currency). We aggregated the municipal data to the micro-region level, and linearly interpolated
the data for the years 1990 and 1995 in order to match our time series of estimated IMR.
9
The availability of data on social programs detailed by micro-region is limited, especially for
those implemented by the Brazilian government before the 1990s. We were able to gather
information on the Family Health Program from the Ministry of Health on-line database. Data
were available since 1998, and included the total number of people and the total number of
children younger than 1 year of age registered in the program. From the Ministry of Social
Development and Fight against Hunger (MDS) website we acquired all information available for
the “Cisternas” and the “Bolsa Familia” CCT programs (Table 1). The total number of water
tanks built in each municipality was obtained for 2003-2006, and the total number of families
receiving the “Bolsa Familia” in each municipality was obtained for 2004-2006. We utilized the
average size of the household by municipality reported in the 2007 Population Count to estimate
the total number of people benefited by each program. We aggregated the data to the microregion level and divided by the total population to obtain measures of program coverage.
METHODS
IMR estimation
Indirect estimations of infant mortality were obtained from information on child survivorship
(Brass 1975b; Brass et al. 1968; Feeney 1980; United Nations 1983). The indirect method allows
the conversion of proportions dead among children ever born to women in each five-year age
group of the reproductive period into estimates of the probability of death, conventionally
reported in life tables as nqx, through the application of multipliers that account for non-mortality
factors impacting the proportions dead (Brass 1975b). These multipliers were estimated based on
the West mortality model (Coale and Demeny 1966) of Trussel’s variant of the original model
(Trussel 1975).
10
The estimates for 0q1 were allocated in time (Brass 1981), resulting in a time series of
estimated IMRs from each data source utilized. That series was subjected to small fluctuations
due to problems in reporting the exact number of children ever born and children surviving. We
addressed this issue by smoothing the series of estimates using a 3-year or 5-year moving
average, depending on the degree of fluctuations. The smoothed series was adjusted by a logistic
function to estimate and project IMRs. The estimation exercise for regions utilized both Census
and PNAD data. The use of PNAD information for 2004-2006 provided recent mortality
estimates that facilitated the adjustment by the logistic (while the Census 2000 allowed us to
obtain indirect estimates allocated until 1998, PNAD data generated estimates until 2004). In the
case of micro-regions, we used only Census data, since the sample design of PNADs does not
allow the disaggregation at the micro-region level (Simões 1999).
We assessed the compatibility between estimates obtained at larger geographical units and
those generated for lower geographical units (regions with Brazil, States with regions, and
micro-regions with States). The basic procedure consisted of obtaining estimates of births
utilizing the P/F ratio method (Brass 1975a; Brass et al. 1968), calculating the number of deaths
by relating the estimations of infant mortality and births, aggregating the number of deaths to the
larger geographical unit, and comparing that number with the deaths obtained by relating the
estimations of infant mortality and births for that larger areal unit. In the rare event of significant
disparities between the two numbers, we preserved the estimations for the smaller geographical
units, and improved the adjustment of the larger areas by the logistic function accordingly
(Simões 1999). In addition, we assessed the sensitivity of the estimations by evaluating the
relationship between the infant mortality in each micro-region and the State, which we expect to
remain roughly constant in the short run. All calculations were performed in Excel (Microsoft,
Seattle, WA, USA), and STATA v.10 (Stata Corp., College Station, TX, USA).
11
Inequality indicators
Inequality in infant mortality was assessed through the use of two indicators. First, we
adapted the concept of inequity ratios proposed by Victora and colleagues (2000) to compute
ratios that compare the extremes in infant mortality, providing a measure of the relative gap in
the regional distribution of IMRs. In other words, we computed the ratio between the smallest
and the largest IMR estimated for each year to evaluate whether or not inequality in infant
mortality was decreasing. Second, we calculated the concentration index (Wagstaff 2000;
Wagstaff, Paci and van Doorslaer 1991; Wagstaff, van Doorslaer and Paci 1989), considering the
micro-region as the unit of analysis, the municipal GDP aggregated to the micro-region as an
income-related indicator, and the infant mortality as the health variable. For each decile of the
GDP distribution we obtained the total number of children aged 0-1 and the average IMR, from
which we estimated the number of deaths. Negative numbers of the index indicate that infant
deaths are concentrated among poor micro-regions, or those with the lowest GDP. All
calculations were performed in Excel and STATA v.10.
Spatial clustering
The presence of spatial clusters of high and low IMR were assessed through the use of a local
indicator of spatial association (LISA), specifically the local Moran’s Ii statistic (Anselin 1995;
Cliff and Ord 1973). Standard normal variates for the statistic, Z(Ii), were computed and the
assessment of significance was based on a normal distribution: significant high values of Z(Ii)
indicate positive spatial autocorrelation – cluster of similar values (either high or low), while
significant low values of Z(Ii) indicate negative spatial autocorrelation – areas of dissimilarity
(Anselin 1995). The weight matrix was defined based on a first-order queen neighborhood
12
definition. All results were corrected for multiple testing utilizing the False Discovery Rate
control procedure (Benjamini and Hochberg 1995; Castro and Singer 2006).
We calculated local Moran’s Ii for each year of available IMR estimates. In addition, we
applied the LISA statistic to the percentage change of IMR observed during each 5-year period
from 1980 to 2005, in order to assess if relative reductions in IMR were spatially randomly
distributed or concentrated in particular areas. Moran’s Ii calculations were performed in
GeoDaTM, FDR corrections were done in Excel, and results were visualized in ArcMap (ESRI,
Redlands, CA, USA).
Change model
We utilized our longitudinal data on estimated infant mortality by micro-regions to assess
how the IMR for each area changed over time, and how the IMR change differed across areas.
Those questions comprise a multilevel model for change, also called a random coefficients
model (Singer and Willett 2002). In the first level of the model we addressed within–microregion changes in infant mortality, and in the second level we appraised between–micro-region
differences in infant mortality change. We collapsed the two levels into one composite model,
assuming a linear functional form:
IMRij = [γ 00 + γ 10TIMEij + γ s 0Wijs + γ 0 k X ik + γ 1k ( X ik ∗ TIMEij )] + [ξ 0i + ξ1iTIMEij + ε ij ]
In this model, IMRij represents the infant mortality for micro-region i at occasion (time) j.
TIME represents years for which we obtained IMR estimates, and is centered on 1980 (the first
year of our time series) so that model estimates represent initial status. Xik represents a set of k
time invariant predictors, Wijs represents a set of s time varying predictors, and the terms in the
second set of brackets represent the composite residual of the model (Singer and Willett 2002).
13
Also, the parameters in the model have specific interpretations: γ 00 and γ 0 k describe inter–
micro-region differences in initial states; γ 10 and γ 1k describe inter–micro-region differences in
infant mortality change; γ 00 and γ 10 represent the average initial status and rate of change,
respectively, of the infant mortality (intercepts of the second level); γ 0 k and γ 1k represent the
effect of each predictor k on the infant mortality change trajectories, and provide a measurement
of increments (or decrements) to average initial status and rate of change, respectively (slopes of
the second level); γ s 0 represents the average difference, over time, in infant mortality due to
predictor s; ε ij represents the residuals of the first level, and indicate the fraction of IMR in
micro-region i that is unpredicted on occasion j; lastly, ξ 0i and ξ1i represent residuals of the
second level, and indicate the portion of initial IMR status and rate of change, respectively, not
explained by the model.
We applied the model described above in two ways. First, we utilized IMRs estimated for
1980, 1985, 1990, 1995, 2000, and 2005 to assess a long-term trajectory of changes in infant
mortality. In this model, we included the following predictors: dummy variables representing the
regions (South was used as reference); a dummy indicating if the micro-region tested significant
for a cluster of high IMR in 1980 (based on the local Moran’s Ii); GDP expressed in the natural
logarithmic scale in order to address the skewness in the distribution, commonly found in
income-related variables; percentage of population living in urban areas, centered around a
constant of 40%; and year specific effects as an attempt to control for unobserved temporal
characteristics that could potentially impact all micro-regions (e.g. policy changes). Both GDP
and urbanization are time-varying predictors. Second, we used a series of estimated IMRs for
1998-2006 to evaluate whether the Health Family Program impacted the trajectory in infant
14
mortality. In addition to the regional dummies, GDP, cluster indicator, and year effects, this
model also included the following time-varying predictors: percentage of the population in the
micro-region covered by the Health Family Program; percentage of children under the age of one
in the micro-region covered by the Health Family Program; and 1-year lagged terms for the
previous two variables. Deviance-based tests and the Akaike Information criterion (AIC) were
used to assess goodness-of-fit (Singer and Willett 2002). All calculations were performed in
STATA v.10.
RESULTS
Indirect estimations of infant mortality: temporal declines and inequalities
We applied the indirect methods of infant mortality estimation to each one of the 557 microregions, to the five regions, and to aggregated country data (Simões 1999). Figure 2 shows the
steps of the estimation for one specific micro-region, named Maceió, located in the Northeast
region. We assembled a time series of IMR estimates for years ending in zero or five for Brazil
and regions (1930-2005), and for micro-regions (1980-2005). Utilizing the year 2000, we
calculated the ratio between the estimated IMR for the micro-region and the estimated IMR for
the corresponding State. We assumed that this relationship would hold roughly constant within a
5-year period, and applied these ratios to the estimated IMR for the State in 2005, generating a
new set of estimated IMRs for micro-regions in 2005. The results did not significantly differ
from those estimated indirectly. All these estimates of infant mortality are currently in use by the
Ministry of Health and by the Pan-American Health Organization (Simões 1999).
The estimated infant mortality for Brazil during 1930-2005, alongside the regions that
consistently recorded the lowest and highest IMRs in the country, South and Northeast regions,
15
respectively, are shown in Figure 3. Brazil had an IMR of 162 in 1930, cut in more than half in
five decades. Considering a threshold of 30 deaths per 1,000 live births (Bourgeois-Pichat 1952),
only two micro-regions were below this mark in 1980, both located in the South region. By the
year 2000, all micro-regions in the South and Center-West were below that threshold. The
Northeast, however, was the only region with IMRs still above 100 per 1,000 in 1975; at 128
infant deaths per 1,000 live births, the region had mortality levels similar to those recorded for
Brazil two decades earlier, despite the significant declines since the 1930s. As a result, only 35%
of the micro-regions in the Northeast had IMR below 30 per 1,000 in 2005.
The decline in infant mortality in the regions with the lowest and highest IMRs did not
significantly shorten the absolute gap in IMR until 1985, when the two curves started to
converge (Figure 3). However, this result per se does not provide an indication that inequalities
between the two regions were also being reducing. In fact, the inequity ratio oscillated around an
average of 1.7 until 1975, jumping to 2.5 in 1980, and reaching a peak of 2.7 in 1985. Although a
steady decline has been observed since 1985, the IMR of the Northeast region in 2005 was still
twice as big as the IMR of the South region, a ratio bigger than that recorded back in 1930.
Considering micro-regions, the absolute gap between the highest and lowest IMR declined
almost 100 points between 1980 and 2005, nonetheless the inequity ratio remained roughly
constant at approximately 6.7 (Figure 4).
A question raised by these results relate to changes in the IMR ranking: Are the declines in
infant mortality barely changing the ranking of micro-regions regarding IMR? Figure 5 suggests
that although fluctuations in the IMR ranking did happen, the top 100 micro-regions (with lowest
IMRs) were restricted to the South and Southeast regions both in 1980 and 2005, while those
occupying rank 400 and above (highest IMRs) were primarily from the Northeast region (with a
16
few from the North region). Considering changes in ranking observed in every 5-year period
during 1980 and 2005, only 55 micro-regions consistently improved their ranking, a third of
them located in the Northeast region. Half of these micro-regions in the Northeast were in Ceará,
a State that implemented a unique package of public health interventions to improve child health
in the late 1980s, recognized in 1993 with UNICEF’s Maurice Pate award (Silva et al. 1999;
Simões and Ortiz 1988; Victora et al. 2000).
Considering the percentile distribution of the GDP, we found that 65% of the micro-regions
within the 10% lowest GDP in 1980 were located in the Northeast region (85% if we considered
both the North and Northeast); only 4% were in the Southeast. Among the 10% highest GDP,
46% of the micro-regions were in the Southeast (65% if we combine the South and Southeast).
While the distribution among the lowest 10% did not change significantly in 2005, the highest
10% became more concentrated in the Southeast, 53% (or 73% combining the South and
Southeast). The health concentration index revealed that poorer micro-regions, as measured by
the GDP, were always in disadvantaged position regarding deaths among infants, when
compared to richer micro-regions (Figure 6). The situation deteriorated between 1980 and 1995,
a period of severe economic instability and fiscal crisis (Silva 1992). After 1995, the index
started to improve, although the concentration index in 2005, -0.11, was not statistically different
from that observed in 1980. Therefore, the recent improvement could be interpreted as a
recovery from 1980s losses.
Spatial clustering of infant mortality
We mapped infant mortality by micro-regions utilizing cutoffs of 30 per 1,000 for low IMR,
and above 50 per 1,000 for high IMR (Figure 7). While the vast majority of micro-regions had
IMRs above 50 in 1980, progressively declines in infant mortality resulted in a complete change
17
of levels in 2005, when the majority had IMRs below 30. A visual inspection of Figure 7
suggests that declines in IMR were steadily observed in all regions, but the Northeast still
concentrated the highest mortality among infants. Indeed, the local Moran’s I revealed that
significant clusters of low IMRs were consistently observed mainly in the South and Southeast
regions of the country, while significant clusters of high IMR were frequently located in the
Northeast (Figure 8).
Therefore, despite the decline in infant mortality during 1980 and 2005, the spatial clustering
pattern of IMRs remained roughly unchanged. Combined with the deterioration of the
concentration index between 1980 and 1995, this finding called for an investigation of the
clustering pattern of the relative rate of change in infant mortality for each 5-year period
included in the analysis. Figure 9 shows a clear geographical divide: between 1980 and 1990,
significant clusters of low decline in IMR were observed in the northern portion of the country,
while the southern portion registered the largest gains. This division progressively weakened,
and the period 2000/2005 recorded a unique pattern and milestone change: all significant clusters
of high infant mortality decline were observed in the Northeast region.
Modeling infant mortality change
We fit 5 different models to assess the long-term trajectory of changes in infant mortality
(Table 1). The variance components of Model 1, an unconditional means model, indicated that
approximately 35% of the variation in IMR was attributable to differences among micro-regions.
Model 2, an unconditional growth model, estimated that 96% of the within–micro-region
variation in infant mortality was associated with changes over time. A correlation coefficient
based on level 2 variance components indicated a strong and negative correlation between the
true IMR rate of change and its level in 1980 (-0.94). In Model 3 we introduced regions as
18
predictors of both initial status and change in infant mortality, and we added year dummies to
control for unobserved temporal effects. Taken together, these variables explained 70% of the
variation in IMR initial status, and 58% of the variation in IMR rates of change. The estimated
IMR initial status for an average micro-region in the South region was 48.2 per 1,000, and 111.3
per 1,000 in the Northeast; the rate of change was 1.3 and 3.2, respectively. Figure 10 shows the
estimated trajectories for each region based on this model, as well as the overall average change
trajectory based on Model 2. Focusing on the period 2000-2005, the curves suggested that while
most regions had an almost flat trajectory, the Northeast region maintained a steep decline in that
period, corroborating the pattern shown in Figure 9.
The explanatory power of the model was improved by further adding the 1980 high cluster
indicator and the GDP: 79% of the variation in IMR initial status, and 69% of the variation in
IMR rate of change were explained by the variables included in Model 4. For each 10% increase
in GDP, the model predicted IMR to decline by an average of 0.006, but this effect was barely
significant (p=0.011). In the last model we added urbanization as a predictor (Model 5), which
was highly significant, indicating that the initial status of infant mortality would be 0.11 points
lower for an average micro-region with 40% of the population living in urban areas.
The models presented in Table 2 investigated IMR changes during a shorter and recent time
period, 1998-2006, utilizing individual year estimates. Model 1 indicated that 92% of the IMR
variation was due to differences among micro-regions. Contrasted with the 35% observed for the
long-term trajectory, this finding suggested that, on average, changes in IMR for each individual
micro-region were less pronounced during 1998-2006, reinforcing the pattern shown in Figure
10, and corroborated by the much lower average change estimated in Model 2 (0.98 per year,
against 2.2 in the long-term change model).
19
Including regional and year dummies (Model 3) explained 67% of the variation in initial
status of infant mortality, and 72% of the variation in rates of change. The initial status of IMR
for an average micro-region in the South and Northeast regions was 19.8 per 1,000 and 44.9 per
1,000, respectively, resulting in an equity ratio of 2.3, the same observed in the long-term
trajectory (Table 1, Model 3). This finding reinforced the pattern observed in Figure 4: the
relative gap between better off and worse off micro-regions did not significantly changed after
the declines in infant mortality.
The last two models attempted to provide an initial assessment of the potential association
between the Family Health Program and changes in infant mortality. Model 4 suggested that the
infant mortality among micro-regions would be reduced as the population covered by the
program increased, but higher as the coverage of children under the age of one increased. The
distribution of these variables indicated that, in 2006, 62% and 24% of the total and younger than
one year population were covered by the program. Regionally, coverage of the total population
in the Northeast was 77%. Both the South and Southeast regions had approximately 50%
coverage. In the case of children younger than one year, the coverage in the South, Southeast and
Northeast regions are the same, about 24%. Therefore, the region with overall highest IMRs did
have the highest coverage of the program considering the total population. However, this clear
distinction was not true for coverage of children under the age of one. The use of 1-year lag
terms (Model 5) did not improve the fit.
DISCUSSION
Our spatio-temporal analysis of infant mortality by micro-regions in Brazil demonstrates that
significant reductions have been observed, particularly since the 1970s, although the changes did
20
not alter significantly the clustering pattern of high and low IMRs in the country. Better
outcomes remained concentrated in the Southern regions, while higher infant mortality was, for
the most part, restricted to the Northeast region. Of special note is the fact that the spatial pattern
of rates of change of infant mortality was dramatically transformed during the 2000-2005 period,
when significant reductions in IMR were clustered in the Northeast. Equally important is the fact
that these declines were not accompanied by significant reductions in inequality of infant
mortality. The relative gap between the better off and the worse off micro-regions remained
unaltered, and therefore infant deaths remained disproportionally concentrated among the
poorest areas. These findings were confirmed by results of the multilevel model for change.
Before we present an in-depth discussion of our findings, a note on data quality is necessary.
Despite improvements in coverage of the registration of vital events, and the development of
new health systems to report deaths, under-reporting of births and deaths is still a problem,
especially in the northern part of Brazil. Therefore, the use of indirect methods of estimation is
the only option to produce reliable indicators of infant mortality, allowing an assessment of
regional disparities, and facilitating the implementation of special programs to target those most
in need. Indeed, IMR estimates produced at different geographical units, regularly provided to
the Ministry of Health (Simões 1999), were the basic information for defining a recently
launched program to reduce infant mortality, which targets 250 municipalities – 154 in the
Northeast region and 96 in the North region (Ministério da Saúde 2009a). In addition, the
government has intensified actions to improve the vital registration system in the northern
portion of the country, utilizing local estimates of the under-reporting of events provided by
these indirect methods.
21
Our results cannot be used as direct evidence of the inverse equity hypothesis, since a
multitude of social, economic, and public health changes and interventions were occurring
concurrently during the study period (Victora et al. 2000). Nonetheless, considering that the IMR
in the most privileged regions of the country was still above 80 per 1,000 in 1970, there was
much room for the richer to take advantage of the services provided through those various
interventions, making it more difficult to reduce the gap between the poor and the rich. That was
specially the case for interventions with national coverage and, therefore, available to any
segment of the population. Although a few targeted programs were also implemented over the
years, a major shift in the social policy agenda took place in mid 1990s, with the introduction of
conditional cash transfer programs, considered to be an effective tool toward poverty reduction
(Barrientos and DeJong 2004, 2006). Although the recent decline in the regional equity ratio
presented in Figure 3 may suggest the success of those programs to effectively reach those most
in need, the roughly flat ratio for micro-regions (Figure 4) indicated that local variabilities at
smaller scales still persisted.
Regarding conditional cash transfer programs, “Bolsa Família” allows poor families to have
access to varied goods (e.g., food) through the transfer of monthly payments, potentially
contributing to reduction in inequalities; indeed, 21% of the decline in the Gini index in Brazil
between 1995 and 2004 was attributed to the “Bolsa Família” program (Soares et al. 2006). In
addition, its conditionalities (e.g., immunization of children, attendance to school, regular health
checks) aim to facilitate the improvement of social and health indicators. While “Progresa”, the
conditional cash transfer program in Mexico, was associated with a 11% decline in infant
mortality among those covered in rural areas (Barham 2006), no formal evaluation of the impact
of “Bolsa Família” on infant mortality in Brazil has been undertaken. Data are available only for
2004-2006. Moreover, since the program unified four pre-existing programs, attempts to utilize a
22
longitudinal model of change (as we applied in this paper) without information prior to 2004
could result in overestimation of the real impacts of the program. The same applies to the
“Cisternas” program, whose impact on infant mortality in Brazil has not been evaluated either.
Until 2008, of the estimated demand of 1,186,601 water tanks – benefiting 4,978,876 people
(d’Alva and Farias 2008), 17% have been constructed. Longitudinal individual-level data,
contrasting recipients and non recipients of the cash transfers and of the water tanks, would be a
more appropriate dataset; to this date, we do not possess such data for infant mortality analysis.
Nonetheless, both the large coverage of the “Bolsa Família” and the scale-up of the “Cisternas”
program are likely to have played important roles on the significant change in the pattern of IMR
decline during 2000-2005, as shown in Figure 9.
Previous studies have shown that, considering the 1993 IMR in Brazil, the impact of the
Family Health Program on infant mortality has been associated with a decline of 5.6% for
municipalities that have been in the program for three years, and 20% for those eight years into
the program (Rocha and Soares 2009). At the State level, increases in the coverage of the Family
Health Program by 10% were associated with an average of 4.6% decline in IMR (Macinko,
Guanais and Souza 2006). Our findings follow the same direction of these previous analyses, and
suggest that data on total population covered by the program may be the most appropriate
indicators for assessment of program relationships with infant mortality.
Several factors commonly associated with declines in infant mortality (Rutstein 2000) did
show improvements recently. The percentage of women that had seven or more prenatal care
consults in the country increased from 22% in 1995 to 56% in 2007. The regional distribution,
however, is very unequal: approximately a third of women in the northern regions, and about
70% in the South and Southeast. Immunization coverage improved from 46% in 1995 to 82% in
23
2007, but achieved a much more equal distribution, with both the North and Northeast regions
registering coverage above 80% in 2007. Equally important were the gains in education: 43% of
women in the reproductive period had only 3 years of schooling in 1980; in 2005 this number
was reduced to 12%. The North and Northeast regions had the highest percentages in 2005, 16%
and 20%, respectively. Nonetheless, the remarkable 70% decline from 1980 levels observed in
these regions was almost certainly a crucial contributing factor for the decline in infant mortality,
considering the largely documented impact of women’s education on IMR (Caldwell 1979).
Another major achievement was the decline in child undernutrition in Brazil since the mid1970s, intensified since the mid-1990s: while the prevalence of stunting declined between 1975
and 1989 at a rate of 5% per year, it was reduced by 6.3% per year between 1996 and 2007
(Monteiro et al. 2009). Although the factors that mostly determined these declines varied over
the years, the most recent decline (from 1996 to 2007) was unique since it was a consequence of
both increases in maternal education and income, and, to a lesser extent, of better access to
health care (Monteiro et al. 2009).
This study has several limitations. First, by using the micro-region as the unit of analysis, we
did not address inequality issues within each micro-region: even the micro-region with highest
IMR and lowest GDP is likely to have spatial variability regarding infant deaths, and differences
in income distribution of the population. Second, we did not include variables on housing
infrastructure (e.g., sanitation, water, and electricity), education, and provision of services (e.g.,
number of doctors per 1,000 people); although often important predictors of infant mortality, due
to limitations of data availability by micro-region. Third, the multilevel models for change that
addressed the impact of the Health Family Program are likely to be subjected to endogeneity
problems. Poor areas often have higher infant mortality and also will draw a large fraction of
candidates to enroll in the program. Therefore, a larger coverage of the Family Health Program
24
may result in infant mortality declines due to better access to health care, but high infant
mortality levels may also result in program coverage increase aiming at improving access to care
and reduce the burden of infant deaths. Which one prevails is yet to be answered. Fourth, since
we did not have individual-level data, we could not assess if reductions in infant mortality were
observed by the poor, who had access to varied programs (e.g., Family Health, “Bolsa Família”,
etc.). Lastly, as it is the case with any ecological analysis, the probability of confounding is non
negligible; we minimized this problem by including year specific variables.
In summary, the decline in infant mortality and the apparent positive signs of reduced
inequalities do send a positive message for the future. The continuation and expansion of current
social programs in the coming years are likely to contribute to the maintenance of trends
demonstrated in our analysis, although two issues need to be addressed. First, it is important to
assess the true causal effects of infant mortality decline, so that current policies may be adapted
and/or improved as needed, and new ones devised. Second, it is crucial to improve understanding
of the determinants of infant mortality (Black, Morris and Bryce 2003; Hill and Pebley 1989;
Mosley and Chen 1984) in the current scenario of lower rates. The year 2010 will be crucial for
both issues, since the country will have presidential elections and a new Demographic Census
will be undertaken. The outcome of the elections will determine how smooth the transition
between the old and the new government will be and, therefore, whether current policies will be
continued or not. The Census will allow revision of the IMR estimates here presented, and
provide much needed individual-level data to comprehensively appraise the determinants of
infant mortality and the causes of the decline here described.
25
REFERENCES
Anselin, L. 1995. "Local Indicators of Spatial Association - LISA." Geographical Analysis
27(2):93-115.
Arbix, G. 2007. "A queda recente da desigualdade no Brasil." Revista Brasileira de Ciências
Sociais 22(64):132-139.
Barham, T. 2006. "Providing a Healthier Start to Life: The Impact of Conditional Cash Transfers
on Neo-Natal and Infant Mortality." Pp. 47: Department of Economics and Institute of
Behavioral Science, University of Colorado Boulder.
Barrientos, A.and J. DeJong. 2004. "Child Poverty and Cash Transfers." Pp. 56 in CHIP Report
No. 4: Childhood Poverty Research and Policy Centre.
—. 2006. "Reducing Child Poverty with Cash Transfers: A Sure Thing?" Development Policy
Review 24(5):537-552.
Barros, R.P., R. Henriques, and R. Mendonça. 2000. "Desigualdade e pobreza no Brasil: retrato
de uma estabilidade inaceitável." Revista Brasileira de Ciências Sociais 15(42):123-142.
Benjamini, Y.and Y. Hochberg. 1995. "Controlling the false discovery rate: a practical and
powerful approach to multiple testing." Journal of the Royal Statistical Society B 57(1):289-300.
Berkman, A., J. Garcia, M. Muñoz-Laboy, V. Paiva, and R. Parker. 2005. "A Critical Analysis of
the Brazilian Response to HIV/AIDS: Lessons Learned for Controlling and Mitigating the
Epidemic in Developing Countries." American Journal of Public Health 95(7):1162-1172.
Black, R.E., S.S. Morris, and J. Bryce. 2003. "Where and why are 10 million children dying
every year?" Lancet 361:2226-2234.
Bourgeois-Pichat, J. 1952. "Essai sur la mortalité biologique de l'homme." Population 7:381394.
Brass, W. 1975a. "Ferility estimates." Pp. 11-17 in Methods for estimating fertility and mortality
from limited and defective data, edited by W. Brass. Chapel Hill, NC: The University of North
Carolina at Chapel Hill, International Program of Laboratories for Population Statistics.
—. 1975b. "Mortality estimates from children ever born and children surviving." Pp. 50-59 in
Methods for estimating fertility and mortality from limited and defective data, edited by W.
Brass. Chapel Hill, NC: The University of North Carolina at Chapel Hill, International Program
of Laboratories for Population Statistics.
—. 1981. "A simple approximation for the time-location of estimates of child mortality from
proportions dead by age of mother." Pp. 6: London School of Hygiene and Tropical Medicine.
Brass, W., A.J. Coale, P. Demeny, D.F. Heisel, F. Lorimer, A. Romaniuk, and E. van de Walle.
1968. The demography of tropical Africa. Princeton, NJ: Princeton University Press.
26
Brito, L.T.L., A.S. Silva, and O.A. d’Alva. 2007. "Avaliação técnica do programa de cisternas no
semi-árido brasileiro." Pp. 199-234 in Avaliação de políticas e programas do MDS: resultados,
edited by J. Vaitsman and R. Paes-Sousa. Brasília, DF: Ministério do Desenvolvimento Social e
Combate à Fome, Secretaria de Avaliação e Gestão da Informação.
Caldwell, J.C. 1979. "Education as a Factor in Mortality Decline An Examination of Nigerian
Data." Population Studies 33(3):395-413.
Cano, W. 1988. "Questão regional e urbanização no desenvolvimento econômico brasileiro pós
1930." Presented at Encontro Nacional de Estudos Populacionais, Olinda, PE.
Carvalho, J.A.M. 1980. "Evolução demográfica recente no Brasil." Pesquisa e Planejamento
Econômico 10(2):527-554.
—. 1997. "Demographic dynamics in Brazil: recent trends and perspectives." Brazilian Journal
of Population Studies 1:5-23.
Castro, J. 1946. Geografia da fome; a fome no Brasil. Rio de Janeiro, RJ: O Cruzeiro.
Castro, M.C. 2001. "Changes in mortality and life expectancy: some methodological issues."
Mathematical Population Studies 9:181-208.
Castro, M.C.and B.H. Singer. 2006. "Controlling the False Discovery Rate: A New Application
to Account for Multiple and Dependent Tests in Local Statistics of Spatial Association."
Geographical Analysis 38(2):180-208.
Cliff, A.D.and J.K. Ord. 1973. Spatial autocorrelation. London: Pion.
Coale, A.J.and P.G. Demeny. 1966. Regional model life tables and stable populations. Princeton,
NJ: Princeton University Press.
d’Alva, O.A.and L.O.P. Farias. 2008. "Programa Cisternas: um estudo sobre a demanda,
cobertura e focalização." Pp. 40 in Cadernos de Estudos Desenvolvimento Social em Debate No.
7. Brasília, DF: Ministério do Desenvolvimento Social e Combate à Fome, Secretaria de
Avaliação e Gestão da Informação.
Demo, P. 1981. Política social na década de 60 e 70. Fortaleza, CE: Edições UFC, Universidade
Federal do Ceará.
Draibe, S. 2003. "A política social no período FHC e o sistema de proteção social." Tempo
Social – USP Novembro:63-101.
Feeney, G. 1980. "Estimating infant mortality trends from child survivorship data." Population
Studies 34(1):109-128.
Frias, L.A.and J.A.M. Carvalho. 1994. "Fecundidade nas regiões brasileiras a partir de 1903:
uma tentativa de reconstrução do passado através das gerações." Presented at Encontro Nacional
de Estudos Populacionais, Caxambú, MG.
27
Giambiagi, F., A.A. Villela, L.B. Castro, and J. Hermann. 2005. Economia brasileira
contemporânea (1945-2004). Rio de Janeiro, RJ: Elsevier: Editora Campus.
Hill, K.and A.R. Pebley. 1989. "Child Mortality in the Developing World." Population and
Development Review 15(4):657-687.
Hoffmann, R. 2000. "Mensuração da desigualdade e da pobreza no Brasil." Pp. 81-107 in
Desigualdade e pobreza no Brasil, edited by R. Henriques. Rio de Janeiro, RJ: IPEA.
IBGE. 1999. Evolução e perspectivas da mortalidade infantil no Brasil. Rio de Janeiro, RJ:
IBGE, Departamento de População e Indicadores Sociais.
Lima, M.H.P., C.M. Rodrigues, J.K.T. Silva, P.C. Martins, S.L. Terron, and R.L.S. Silva. 2002.
Divisão Territorial Brasileira. Rio de Janeiro, RJ: IBGE, Diretoria de Geociências,
Departamento de Geografia, Departamento de Estruturas Territoriais.
Lindert, K., A. Linder, J. Hobbs, and B. Brière. 2007. "The Nuts and Bolts of Brazil’s Bolsa
Família Program: Implementing Conditional Cash Transfers in a Decentralized Context." Pp.
145 in Social Protection Discussion Paper no. 0709: The World Bank.
Macinko, J., F.C. Guanais, and M.F.M. Souza. 2006. "Evaluation of the impact of the Family
Health Program on infant mortality in Brazil, 1990–2002." Journal of Epidemiology and
Community Health 60:13-19.
Madeira, J.L.and C.C.S. Simões. 1972. "Estimativas preliminares da população urbana e rural
segundo as Unidades da Federação, de 1960/1980 por uma nova metodologia." Revista
Brasileira de Estatística 33(129):3-11.
Marques, R.M. 2005. "A Importância do Bolsa Família nos Municípios Brasileiros." Pp. 40 in
Cadernos de Estudos Desenvolvimento Social em Debate No. 1. Brasília, DF: Ministério do
Desenvolvimento Social e Combate à Fome, Secretaria de Avaliação e Gestão da Informação.
Medici, A.C.and A.C. Agune. 1994. "Desigualdades sociais e desenvolvimento no Brasil: uma
análise dos anos 80 ao nível regional." Ensaios 15(2):458-488.
Ministério da Integração Nacional. 2005. Nova Delimitação do Semi-Árido Brasileiro. Brasília,
DF: Ministério da Integração Nacional, Secretaria de Políticas de Desenvolvimento Regional.
Ministério da Saúde. 2008. Pesquisa Nacional de Demografia e Saúde da Mulher e da Criança,
PNDS 2006. Relatório Final. Brasília, DF: Ministério da Saúde.
—. 2009a. "Pacto pela Redução da Mortalidade Infantil Nordeste-Amazônia Legal." Brasília,
DF: Ministério da Saúde.
http://portal.saude.gov.br/portal/saude/profissional/area.cfm?id_area=1583 (Accessed in August,
2009).
—. 2009b. "Saúde da Família." Brasília, DF: Ministério da Saúde.
http://dtr2004.saude.gov.br/dab/index.php (Accessed in August, 2009).
28
Monteiro, C.A., M.H.D.A. Benicio, S.C. Konno, A.C.F. Silva, A.L.L. Lima, and W.L. Conde.
2009. "Causes for the decline in child under-nutrition in Brazil, 1996-2007." Revista de Saúde
Pública 43(1):1-8.
Mosley, W.and L. Chen. 1984. "An analytic framework for the study of child survival in
developing countries." Population and Development Review 10:25-45.
Nunn, A.S., E.M. Fonseca, F.I. Bastos, and S. Gruskin. 2009. "AIDSTreatment In Brazil:
Impacts and Challenges." Health Affairs 28(4):1103-1113.
Pereira, L.C.B. 1994. "A economia e a política do Plano Real." Revista de Economia Política
14(4):129-149.
Pero, V.and D. Szerman. 2005. "The New Generation of Social Programs in Brazil." Pp. 28 in
Série Seminários de Pesquisa. Rio de Janeiro, RJ: Universidade Federal do Rio de Janeiro,
Instituto de Economia.
Pinto, L.F.S. 2002. Gestão-cidadã: ações estratégicas para a participação social no Brasil. Rio
de Janeiro, RJ: FGV Editora.
Rocha, R.and R.R. Soares. 2009. "Evaluating the Impact of Community-Based Health
Interventions: Evidence from Brazil’s Family Health Program." Pp. 60 in Discussion Paper No.
4119. Bonn: Institute for the Study of Labor (IZA).
Rocha, S. 1998. "Desigualdade Regional e Pobreza no Brasil: a Evolução - 1981/95." Pp. 27 in
Texto para Discussão Nº 567. Rio de Janeiro, RJ.
Rutstein, S.O. 2000. "Factors associated with trends in infant and child mortality in developing
countries during the 1990s." Bulletin of the World Health Organization 78(10):1256-1270.
Santos, S.M.C.and L.M.P. Santos. 2007. "Avaliação de políticas públicas de segurança alimentar
e combate à fome no período de 1995-2002. 1 – Abordagem metodológica." Cadernos de Saúde
Pública 23(5):1029-1040.
Silva, A.C., A.M. Leite, F.M.O. Andrade, J.S. Campos, L.R.S.K. Oliveira, and R.L.P. Diniz.
1999. Viva Criança: Os caminhos da sobrevivência infantil no Ceará. Fortaleza, CE: Fundação
Demócrito Rocha.
Silva, L.C.E. 1992. "O que mostram os indicadores sobre a pobreza na década perdida." Brasília,
DF: Instituto de Pesquisa Econômica Aplicada - IPEA. Textos para Discussão 274.
Simões, C.C.S. 1997. "A Mortalidade infantil na transição da mortalidade no Brasil: um estudo
comparativo entre o Nordeste e o Sudeste." Departamento de Demografia, CEDEPLAR, UFMG.
—. 1999. Estimativa da Mortalidade Infantil por Microrregiões e Municípios. Brasília, DF:
Ministério da Saúde, e Organização Panamericana de Saúde.
—. 2002. Perfis de saúde e de mortalidade no Brasil: uma análise de seus condicionantes em
grupos populacionais específicos. Brasília, DF: Organização Pan-Americana da Saúde.
29
—. 2006. A transição da fecundidade no Brasil: análise de seus determinantes e as novas
questões demográficas. São Paulo, SP: Arbeit Factory.
Simões, C.C.S.and L.P. Ortiz. 1988. "A mortalidade infantil no Brasil nos anos 80." Pp. 243-268
in Crise e infância no Brasil: o impacto das políticas de ajustamento econômico, edited by J.P.Z.
Chadad and R. Cervini. São Paulo, SP: IPE/USP.
Singer, J.D.and J.B. Willett. 2002. Applied longitudinal data analysis: modeling change and
event occurrence. Oxford; New York: Oxford University Press.
Soares, F.V., S. Soares, M. Medeiros, and R.G. Osório. 2006. "Cash Transfer Programmes in
Brazil: Impacts on Inequality and Poverty." Pp. 35 in Working Paper No. 21. Brasília, DF:
United Nations Development Programme, International Poverty Centre.
Tavares, L. 1981. "Política social e pobreza: o Brasil no contexto latino-americano." Rio de
Janeiro, RJ: UFRJ.
Trussel, J. 1975. "A re-estimation of the multiplying factors for the Brass technique for
determining childhood survivorship rates." Population Studies 29(1):97-107.
UNICEF. 1990. "A promise to children." New York: UNICEF (http://www.unicef.org/wsc/).
—. 2005. Tracking Progress in Child Survival: The 2005 Report. New York: The United Nations
Children’s Fund (UNICEF).
United Nations. 1983. Manual X: Indirect techniques for demographic estimation. New York,
NY: Population Division of the Department of International Economic and Social Affairs,
Committee on Population and Demography of the National Research Council.
Vasconcelos, F.A.G. 2005. "Combate à fome no Brasil: uma análise histórica de Vargas a Lula."
Revista de Nutrição 18(4):439-457.
Victora, C.G., J.P. Vaughan, F.C. Barros, A.C. Silva, and E. Tomasi. 2000. "Explaining trends in
inequities: evidence from Brazilian child health studies." Lancet 356:1093-1098.
Wagstaff, A. 2000. "Socioeconomic inequalities in child mortality: comparisons across nine
developing countries." Bulletin of the World Health Organization 78(1):19-29.
Wagstaff, A., P. Paci, and E. van Doorslaer. 1991. "On the measurement of inequalities in
health." Social Science and Medicine 33(5):545-557.
Wagstaff, A., E. van Doorslaer, and P. Paci. 1989. "Equity in the finance and delivery of health
care: some tentative cross-country comparisons." Oxford Review of Economic Policy 5:89-112.
World Bank. 1999. "World Development Indicators 1998/99." Washington, DC: World Bank.
—. 2007. "World Development Indicators 2007." Washington, DC: World Bank.
30
Figure 1 – Delineation of the semi-arid region in Brazil and selected administrative
divisions (regions, micro-regions and municipalities)
31
Figure 2 – Estimated infant mortality for Maceió micro-region – 1970-2005
140
IMR (per 1,000 live births)
120
100
80
60
40
20
0
1965
1970
1975
1980
1985
1990
1995
2000
2005
Year
Indirect estimates
Smoothed (moving average)
Adjusted (Logistic)
Source: Indirect estimations produced by the authors (details provided in Methods section).
32
Figure 3 – Infant mortality rates, Brazil and regions with consistently lower
and higher rates in the country – 1930-2005
200
3.0
180
2.5
140
2.0
120
1.5
100
80
Inequity ratio
IMR (per 1,000 live births)
160
1.0
60
40
0.5
20
0
0.0
1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Year
Brazil
Highest Regional IMR (Northeast)
Lowest Regional IMR (South)
Inequity ratio
Source: Indirect estimations produced by the authors (details provided in Methods section).
33
160
8
140
7
120
6
100
5
80
4
60
3
40
2
20
1
0
0
1980
1985
1990
Absolute gap
1995
2000
Inequity ratio
Difference between smallest and largest IMR
Figure 4 – Equity ratios and absolute difference between the smallest and largest IMR
estimated for micro-regions – 1980-2005
2005
Relative gap
Source: Indirect estimations produced by the authors (details provided in Methods section).
34
Figure 5 – Changes in the ranking of micro-regions according to estimated infant mortality
(rank 1 = lowest IMR) – 1980 and 2005
600
North
Northeast
Center-West
Southeast
South
500
Rank in 2005
400
300
200
100
0
0
100
200
300
400
500
600
Rank in 1980
Source: Indirect estimations produced by the authors (details provided in Methods section).
35
Figure 6 – Concentration Index of infant deaths by micro-region GDP – 1980-2005
-0.09
-0.10
Concentration Index
-0.11
-0.12
-0.13
-0.14
-0.15
-0.16
1980
1985
1990
1995
2000
2005
Year
36
Figure 7 – Infant mortality per 1,000 live births by micro-region – 1980-2005
Source: Indirect estimations produced by the authors (details provided in Methods section).
37
Figure 8 – Summary of clustering pattern of infant mortality by micro-region and based on
estimated IMR for 1980, 1985, 1990, 1995, 2000 and 2005
38
Figure 9 – Clustering pattern of percentage changes in infant mortality by micro-region for
each 5-year period between 1980 and 2005
39
20
Linear predictor, fixed portion
40
60
80
100
120
Figure 10 – Trajectories of change in infant mortality by region based on a multilevel
model that control a year specific effect (Model 3). The average trajectory is based on the
unconditional growth model (Model 2).
1980
1985
1990
1995
2000
2005
year
Norte
Nordeste
Sul
Sudeste
CO
Average change trajectory
40
Table 1 – Models of initial status and change in infant mortality rates
by micro-regions – 1980, 1985, 1990, 1995, 2000 and 2005
Fixed effects
Rate of change
Intercept
1985
1990
1995
2000
North
Northeast
Southeast
Center-West
Cluster high IMR, 1980
ln(GDP)
Urbanization
Intercept
North
Northeast
Southeast
Center-West
Cluster high IMR, 1980
Variance components:
Level 1
within
Level 2
rate of change
initial status
covariance
γ00
γ01
γ02
γ03
γ04
γ05
γ06
γ07
γ08
γ09
γ20
γ30
γ10
γ15
γ16
γ17
γ18
γ19
2
σε
2
σ1
2
σ0
σ01
Mod. 1
45.931
532.384
291.133
Mod. 2
73.314
Mod. 3
48.214
-0.551
-4.482
-7.513
-5.941
21.641
63.083
7.086
11.585
Mod. 4
58.259
-0.511
-4.472
-7.656
-5.962
20.934
39.689
7.085
11.069
33.645
ns
-0.492
-2.191
-1.272
-0.538
-1.906
-0.247
-0.382
-1.262
-0.530
-1.064
-0.244
-0.370
-1.243
Mod. 5
48.143
-0.556
-4.483
-7.496
-5.938
20.307
39.095
8.584
12.003
33.782
ns
0.061
-0.107
-1.171
-0.531
-1.069
-0.273
-0.371
-1.252
22.022
1.034
996.578
-31.277
9.546
0.431
302.776
-10.633
9.564
0.317
215.598
-7.494
9.545
0.318
214.484
-7.516
23121.3
23133.3
20537.1
20573.1
20347.9
20389.9
20322.9
20366.9
Goodness-of-fit:
Deviance
AIC
Note:
ns
31217.0
31223.0
= not significant. The year 2005 was removed due to collinearity.
41
Table 2 – Models of initial status and change in infant mortality rates
by micro-regions – 1998-2006
Fixed effects
Rate of change
Intercept
1999
2000
2001
2002
2003
2004
2005
North
Northeast
Southeast
Center-West
Cluster high IMR, 1980
Urbanization
% coverage HF-T
% coverage HF-C
lag % coverage HF-T
lag % coverage HF-C
γ00
γ01
γ02
γ03
γ04
γ05
γ06
γ07
γ08
γ09
γ010
γ011
γ012
γ20
γ40
γ60
γ80
γ100
Intercept
North
Northeast
Southeast
Center-West
Cluster high IMR, 1980
γ10
γ18
γ19
γ110
γ111
γ112
Variance components:
Level 1
within
Level 2
rate of change
initial status
covariance
2
σε
2
σ1
2
σ0
σ01
Mod. 1
25.748
Mod. 2
29.665
Mod. 3
19.848
-0.571
-0.942
-1.125
-1.154
-1.038
-0.797
-0.446
10.449
25.088
1.955
3.698
Mod. 4
20.679
-0.585
-0.933
-1.080
-1.099
-1.006
-0.769
-0.437
10.059
18.813
ns
2.295
3.874
8.721
-0.033
-0.008
0.007
Mod. 5
20.658
-0.599
-0.979
-1.136
-1.130
-1.004
-0.781
-0.434
10.073
18.813
ns
2.286
3.854
8.691
-0.032
-0.006
0.003
9.865
120.641
-0.979
-0.555
-0.312
-1.097
-0.079
-0.100
-0.492
-0.311
-0.823
ns
-0.091
ns
-0.080
-0.379
-0.495
-0.315
-0.828
ns
-0.089
ns
-0.079
-0.374
0.334
0.312
171.938
-6.903
0.103
0.088
56.557
-1.815
0.099
0.076
49.306
-1.510
0.100
0.076
49.371
-1.518
14474.7
14486.7
9153.1
9191.1
8932.0
8980.0
8972.0
9020.0
Goodness-of-fit:
Deviance
AIC
28273.5
28279.5
Note: ns = not significant. The year 2006 was removed due to collinearity.
42